Ensemble Trend Classification in the Foreign Exchange Market Using Class Variable Fitting

Andrew Kreimer, Maya Herman

Abstract: We present a method for ensemble classification of trends in the foreign exchange market using historical data, technical analysis and class variable fitting. We have implemented a complete closed source algorithmic trading platform in Java and MQL. In contradiction to standard concrete price prediction or trend classification, we apply ensemble trend classification and search for optimal class variable. We use single timeframe in contradiction to multiple timeframes analysis approach. We show substantial profitably applying the trading strategies derived by our approach. This paper has two main objectives. The first, to present a new trend definition by expanding the search space for more efficient trading strategies. The second, is to present a new algorithmic trading platform and provide a live trading historical performance rather than back testing results. While previous works in the field tend to incorporate single trading strategy, we show a method for finding multiple trading strategies for various assets.